This released repository implements Quantifying Meibomian Gland Morphology Using Artificial Intelligence. More specifically, the gland segmentation of meibography images.
- minoconda3
- pytorch==1.11.0
- torchvision==0.12.0
- use pip install if you miss any dependencies
We are not releasing the training data to the public at this moment. However we are providing a sample set that you can run and validate results. See sample set here.
To download the trained model, please fill this form. Note that we will provide you with both upper lid and lower lid segmentation model.
If you want to work on your own data, change data_path
and raw_dir
in create_annotations.py accordingly.
Then, run the following:
python data/scripts/create_annotations.py
We are training upper and lower lid segmentation separately.
Run the following:
cd code
python train.py --dataset meibo --gpu YOUR_GPU_NUMBER
Run the following:
cd code
python pred_list.py --lst val.lst --usegpu --dataset meibo --model PATH_TO_YOUR_MODEL
See some sample visualization results here.
The use of this software is released under CC BY-NC-SA 4.0.
Please cite our paper if you find this repo useful.
@article{wang2021quantifying,
title={Quantifying meibomian gland morphology using artificial intelligence},
author={Wang, Jiayun and Li, Shixuan and Yeh, Thao N and Chakraborty, Rudrasis and Graham, Andrew D and Stella, X Yu and Lin, Meng C},
journal={Optometry and Vision Science},
volume={98},
number={9},
pages={1094--1103},
year={2021},
publisher={LWW}
}
This repo is heavily based on instance-segmentation-pytorch.